Renew or Churn Call
Health scores drift into vibes because nobody defines the levels. This skill fixes five levels — from will churn to will renew and expand — each with the evidence that puts an account there, and returns a probability for every level. The next move is mapped to the level, so the output is an action, not a colour.
Where this sits
After cs-health-scorecard (which gathers the dimensions) and before renewal-playbook or cs-escalation-brief (which run the move). This one decides which move.
What This Skill Produces
- The account state — six fields, filled from CRM and usage evidence
- The renewal score — one of five defined levels, with a probability per level and a confidence score
- The next move — mapped from the level (save play, playbook, discovery call, standard motion, expansion)
- The flip signal — the one change in state that would move the level
- The forecast line — for the renewals sheet: level, probability, move, owner, date
Required Inputs
Ask for these if not provided; mark unknowns rather than guessing:
- account — name, ARR, renewal date, term
- usage — active seats vs licensed, 90-day trend, key feature adoption
- engagement — last executive touch, QBR held, support tickets open/closed, NPS or CSAT
- champion — still in role? sponsor aligned?
- commercial — price change at renewal, competitor in play, procurement signals
- history — prior escalations, credits, missed commitments
Framework: Five Defined Levels
| Level |
Evidence that puts an account here |
Next move |
| 0 · will churn |
Champion gone, usage collapsing, competitor selected, or procurement told us |
Escalate now: exec-to-exec call, save play with a concession budget |
| 1 · at risk |
Usage down, exec contact stale, open escalation or price objection |
Run the renewal playbook with a 60-day save plan |
| 2 · uncertain |
Mixed signals: healthy usage but no sponsor, or engaged but under-adopted |
Discovery call to find the sponsor and the value gap |
| 3 · likely renew |
Usage steady, sponsor engaged, no blockers, price accepted |
Standard renewal motion; confirm timeline with procurement |
| 4 · renew and expand |
Growing usage, seats capped, expansion asked for |
Expansion proposal before the renewal paperwork |
Thresholds. Act on the mapped move when confidence ≥ 0.55 and the top level's probability ≥ 0.6. Otherwise hold and look at the shape: a 0.40 / 0.40 split between at risk and likely renew is a discovery call, not a save play — the split itself says the sponsor question is unanswered.
Getting probabilities. Any calibrated decision model that scores against ordered, described levels serves this contract; the repo ships an adapter (node integrations/jev/decide.mjs renew-or-churn-call --state state.json, contract in integrations/jev/decisions/). Without one, the assistant estimates and labels it.
Output Format
Renewal call: [account] · ARR [x] · renews [date]
State (six rows, unknown where unknown)
Score
| Level |
Probability |
| 0 will churn |
0.xx |
| 1 at risk |
0.xx |
| 2 uncertain |
0.xx |
| 3 likely renew |
0.xx |
| 4 renew and expand |
0.xx |
Expected level: x.x · Confidence: 0.xx · Source: [model / estimate]
Verdict: [act: move] or [hold — shape says: …]
Flip signal: [the one state change that moves the level, and who would know]
Forecast line: [account] · L[n] p=0.xx · [move] · [owner] · [date]
Quality Checks
Anti-Patterns
- RAG colours with no definition — "amber" that means something different to each CSM
- Averaging a bimodal distribution into "medium risk"
- Save play by reflex — running concessions on an account whose problem is a missing sponsor
- Ignoring the champion field — the single most predictive signal, most often left blank
- Letting the score replace the call — the account owner owns the forecast; this is the evidence
Example Trigger Phrases
- "Will Acme renew in November? Give me a real read."
- "Churn risk on this account — save play or not?"
- "Score this renewal and tell me the next move."
1---2name: renew-or-churn-call3description: "Score an account's renewal likelihood on a five-level scale with probabilities, and name the next move for the account owner. Use when asked will this account renew, what's the churn risk on this account, should we run a save play, or when a renewal is 30–120 days out and the owner needs a calibrated read rather than a gut feel. Produces the account state in a fixed schema, the five-level renewal score with a probability per level and confidence, the mapped next move, and the single signal that would change the call."4version: 1.0.05disable-model-invocation: true # skill-optimizer6metadata:7 source: https://skillmd.com/skills/slipcast-dev/renew-or-churn-call8---910# Renew or Churn Call1112Health scores drift into vibes because nobody defines the levels. This skill fixes five levels — from *will churn* to *will renew and expand* — each with the evidence that puts an account there, and returns a probability for every level. The next move is mapped to the level, so the output is an action, not a colour.1314## Where this sits1516After [`cs-health-scorecard`](../cs-health-scorecard/SKILL.md) (which gathers the dimensions) and before [`renewal-playbook`](../renewal-playbook/SKILL.md) or [`cs-escalation-brief`](../cs-escalation-brief/SKILL.md) (which run the move). This one decides which move.1718## What This Skill Produces1920- **The account state** — six fields, filled from CRM and usage evidence21- **The renewal score** — one of five defined levels, with a probability per level and a confidence score22- **The next move** — mapped from the level (save play, playbook, discovery call, standard motion, expansion)23- **The flip signal** — the one change in state that would move the level24- **The forecast line** — for the renewals sheet: level, probability, move, owner, date2526## Required Inputs2728Ask for these if not provided; mark unknowns rather than guessing:29- **account** — name, ARR, renewal date, term30- **usage** — active seats vs licensed, 90-day trend, key feature adoption31- **engagement** — last executive touch, QBR held, support tickets open/closed, NPS or CSAT32- **champion** — still in role? sponsor aligned?33- **commercial** — price change at renewal, competitor in play, procurement signals34- **history** — prior escalations, credits, missed commitments3536## Framework: Five Defined Levels3738| Level | Evidence that puts an account here | Next move |39|---|---|---|40| 0 · will churn | Champion gone, usage collapsing, competitor selected, or procurement told us | Escalate now: exec-to-exec call, save play with a concession budget |41| 1 · at risk | Usage down, exec contact stale, open escalation or price objection | Run the renewal playbook with a 60-day save plan |42| 2 · uncertain | Mixed signals: healthy usage but no sponsor, or engaged but under-adopted | Discovery call to find the sponsor and the value gap |43| 3 · likely renew | Usage steady, sponsor engaged, no blockers, price accepted | Standard renewal motion; confirm timeline with procurement |44| 4 · renew and expand | Growing usage, seats capped, expansion asked for | Expansion proposal before the renewal paperwork |4546**Thresholds.** Act on the mapped move when confidence ≥ 0.55 and the top level's probability ≥ 0.6. Otherwise **hold and look at the shape**: a 0.40 / 0.40 split between *at risk* and *likely renew* is a discovery call, not a save play — the split itself says the sponsor question is unanswered.4748**Getting probabilities.** Any calibrated decision model that scores against ordered, described levels serves this contract; the repo ships an adapter (`node integrations/jev/decide.mjs renew-or-churn-call --state state.json`, contract in `integrations/jev/decisions/`). Without one, the assistant estimates and labels it.4950## Output Format5152### Renewal call: [account] · ARR [x] · renews [date]5354**State** (six rows, `unknown` where unknown)5556**Score**57| Level | Probability |58|---|---|59| 0 will churn | 0.xx |60| 1 at risk | 0.xx |61| 2 uncertain | 0.xx |62| 3 likely renew | 0.xx |63| 4 renew and expand | 0.xx |6465Expected level: x.x · Confidence: 0.xx · Source: [model / estimate]6667**Verdict:** **[act: move]** or **[hold — shape says: …]**68**Flip signal:** [the one state change that moves the level, and who would know]69**Forecast line:** `[account] · L[n] p=0.xx · [move] · [owner] · [date]`7071## Quality Checks72- [ ] Every state field is evidence or `unknown`; no invented usage numbers73- [ ] All five probabilities shown; expected level computed74- [ ] The move is the one mapped to the level, not a freelance idea75- [ ] A split distribution is read as a question, not averaged into a colour76- [ ] The forecast line is one line and fits the renewals sheet7778## Anti-Patterns79- **RAG colours with no definition** — "amber" that means something different to each CSM80- **Averaging a bimodal distribution** into "medium risk"81- **Save play by reflex** — running concessions on an account whose problem is a missing sponsor82- **Ignoring the champion field** — the single most predictive signal, most often left blank83- **Letting the score replace the call** — the account owner owns the forecast; this is the evidence8485## Example Trigger Phrases86- "Will Acme renew in November? Give me a real read."87- "Churn risk on this account — save play or not?"88- "Score this renewal and tell me the next move."